According to VentureBeat, Anthropic today unveiled Claude Opus 5.5, its new model aimed at long-term coding, research and professional work. This model not only outperforms, but also offers a lower price compared to Anthropic's previous flagship models Mythos 5.1 and Fable 5.1, released just a few weeks ago. Opus 5.5 is priced via the API at $4 per million input tokens and $20 per million output tokens, which is 20% lower than Opus 5. Typical workloads are said to cost about 40% less because the model uses fewer tokens to complete tasks. The company also reduced cache write prices by 20%, from $6.25 to $5 per million tokens, and cache read prices were reduced even more significantly, from $0.50 to $0.20 per million tokens. At the same time, Opus 5.5 achieved some of the highest scores of all Anthropic models, including coding (Terminal-Bench 4.0, FrontierCode v1.1 Main, CursorBench 4.0), knowledge work (GDPval-AA v2.1) and scientific multidisciplinary thinking (Terminal-Bench-Science 0.1³, Humanity's Last Exam). Opus 5.5 significantly exceeds
The performance of Fable 5.1, the previous flagship for a wide audience, is shattering, although Fable 5.1 costs 150% more in terms of API. The model's release comes amid growing competition among model vendors who are increasingly focused on the amount of useful work done per dollar, rather than just test results. Opus 5.5 is positioned as the most efficient model for programming that requires an agent to work with code for extended periods of time, and not just answer individual coding questions. The company reports achieving 66.4% on Terminal-Bench 4.0, while Fable 5.1 achieved 55.8% and Opus 5 52.3%. On FrontierCode v1.1, Main Opus 5.5 achieved 54.4%, compared to 50.3% for Fable 5.1 and 48.0% for Opus 5. It's impressive that Opus 5.5 achieves these results while remaining 60% cheaper at base API token prices. Anthropic also points to large real-world workloads. One early tester completed migrating 680,000 lines of code in less than a day, while another used Opus 5.5 to audit and fix a 200,000-line codebase in less than three hours, while Opus 5 took over 20 hours and 2.5 times.
for more tokens. Opus 5.5 used "some of the lowest token counts and steps" that GitHub measured in Copilot CLI and VS Code testing, while solving more terminal problems than Opus 5 in less than half the steps, according to GitHub Chief Product Officer Mario Rodriguez. Other early testers, including Lovable, Spotify, and Optiver, also reported reductions in hits, output tokens, or execution time. These results were selected by vendors, not independent evaluations, but they indicate an operational metric that Anthropic wants businesses to track: how much orchestration, retries, and token consumption occurs before a job is completed. Also today, OpenAI released the GPT-6 Sol and GPT-6 Luna, two lower-priced models from the GPT-6 family. Sol costs $2 per million input tokens and $10 per million output tokens—exactly half the base prices of Opus 5.5. Luna drops to just $0.10/$0.50, a 97.5% reduction on both counts compared to Opus 5.5. These launches raise an important question for businesses: not which model is the most powerful, but how much useful battery life each model has
can be done for a certain amount of money. Opus 5.5 has a context window of 1 million tokens, supports up to 128,000 output tokens synchronously, and uses the API model identifier claude-opus-5-5. Anthropic also supports output of up to 300,000 tokens through its Message Batches API in beta. However, Opus 5.5 is not a direct replacement for all Opus 5 applications. Coding and other professional workflows also require cost-saving models. Anthropic also markets Opus 5.5 for financial analysis, research, document preparation, and other professional work. On GDPval-AA v2.1, a performance benchmark across 44 professions, Anthropic reports 1,846 Elo for Opus 5.5, compared to 1,735 for Fable 5.1 and 1,708 for Opus 5. Anthropic reports that Opus 5.5 performs better than recent Claude models on almost all measures in its automated behavior audit. The new containment assessment reports that Opus 5.5 tried to bypass boundaries approximately 85% less often than Opus 5 or Claude Mythos 5.1. However, the company also cautions that reliably detecting every bug before deployment remains an ongoing challenge. Claude Opus 5.5 d
Available through the Anthropic API and the AWS, Google Cloud and Microsoft ecosystems. GPT-6 Sol and Luna are available through the OpenAI API and are being gradually implemented through ChatGPT Work and Codex. The competition after the release of both models becomes more complex than just ranking based on test results. Anthropic has significantly reduced the price of Claude's high-end capabilities: Opus 5.5 beats the previous flagship on most metrics reported by Anthropic, while costing 60% less at base API token prices. The material was prepared by a neural network based on the original publication. AI may make mistakes in details (numbers, names, dates, quotes) - for important information, check with the original source. Anthropic introduced Claude Opus 5.5 with high performance and low cost OpenAI introduced the GPT-6 Sol and Luna models, reducing API prices by 50% Kyutai released the Voice of Reason model for verbal solving of mathematical problems. Nokia opens access to AnyJev, a library for decision-making without training. OpenAI introduced GPT-6 Sol and Luna with new API pricing, reducing the cost by 50%.
